Methods and apparatuses are described for automated computer text classification and routing using artificial intelligence transfer learning. A server trains a word embedding model using one-hot vectors of word pairs from a filtered first corpus of unstructured computer text and a filtered second corpus of unstructured computer text, using an artificial intelligence neural network. The server trains a long short-term memory model using vector matrices that correspond to sentences in the filtered second corpus of unstructured computer text, and labels. The server receives a message, generates a matrix for each sentence in the message by applying the trained word embedding model, generates one or more labels, and a probability for each label, for each sentence in the message by applying the trained long short-term memory model, and routes the message to a second client computing device based upon an assigned label.
Legal claims defining the scope of protection, as filed with the USPTO.
1. A system used in a computing environment in which unstructured computer text is analyzed for classification and routing using artificial intelligence transfer learning, the system comprising: a computer data store including a first corpus of unstructured computer text associated with a first domain and a second corpus of unstructured computer text associated with a second domain; a server computing device in communication with the computer data store, the server computing device programmed to: filter the first corpus of unstructured computer text and the second corpus of unstructured text using natural language processing; generate a first vocabulary data set based upon the filtered first corpus of unstructured computer text; generate a second vocabulary data set based upon the filtered second corpus of unstructured computer text; generate a one-hot vector for each word in the first vocabulary data set and the second vocabulary data set; train, using an artificial intelligence neural network executing on the server computing device, a word embedding model using the one-hot vectors of word pairs from the filtered first corpus of unstructured computer text and the filtered second corpus of unstructured computer text; create a multidimensional vector for each word in the filtered first corpus of unstructured computer text and the filtered second corpus of unstructured computer text using the trained word embedding model; build a 3D tensor for the filtered second corpus of unstructured computer text using the trained word embedding model, the 3D tensor comprising a plurality of matrices, wherein each matrix corresponds to a sentence in the filtered second corpus of unstructured computer text and each matrix comprises a plurality of vectors, each vector corresponding to a word in the sentence; apply one or more labels to the plurality of matrices in the 3D tensor; train, using a recurrent artificial intelligence neural network executing on the server computing device, a long short-term memory model using the plurality of matrices in the 3D tensor and the corresponding labels; and a first client computing device, in communication with the server computing device, that generates a message comprising one or more sentences of unstructured computer text associated with the second domain; wherein the server computing device: filters the unstructured computer text in the message using natural language processing; generates a matrix for each sentence in the unstructured computer text in the message by applying the trained word embedding model to each word in the sentence; identifies one or more labels, and a probability for each label, for each sentence in the unstructured computer text in the message by applying the trained long short-term memory model to the generated matrix for each sentence; assigns a label having a highest probability to each sentence in the unstructured computer text; and routes the message to a second client computing device based upon the assigned label.
2. The system of claim 1 , wherein filtering the first corpus of unstructured computer text and the second corpus of unstructured text using natural language processing comprises one or more of: (i) removing stopwords from the first corpus of unstructured computer text and the second corpus of unstructured text, (ii) lemmatizing the first corpus of unstructured computer text and the second corpus of unstructured text, or (iii) removing one or more symbols or digits from the first corpus of unstructured computer text and the second corpus of unstructured text.
3. The system of claim 1 , wherein the first corpus of unstructured computer text and the second corpus of unstructured computer text are input into the computer data store via a web page, input directly into the computer data store via a first computer file, input into the computer data store via a data feed, or any combination thereof.
4. The system of claim 1 , wherein generating a first vocabulary data set based upon the filtered first corpus of unstructured computer text comprises, for each word in the filtered first corpus of unstructured computer text: determining a number of times that the word appears in the filtered first corpus of unstructured computer text, and adding the word to a first vocabulary data set if the number of times that the word appears in the filtered first corpus of unstructured computer text is above a predetermined threshold.
5. The system of claim 1 , wherein generating a second vocabulary data set based upon the filtered second corpus of unstructured computer text comprises, for each word in the filtered second corpus of unstructured computer text: determining a number of times that the word appears in the filtered second corpus of unstructured computer text; and adding the word to a second vocabulary data set if the number of times that the word appears in the filtered second corpus of unstructured computer text is above a predetermined threshold.
6. The system of claim 1 , wherein the artificial intelligence neural network that trains the word embedding model comprises a shallow neural network having an input layer, a hidden layer, and an output layer.
7. The system of claim 1 , wherein the first corpus of unstructured computer text associated with the first domain is larger than the second corpus of unstructured computer text associated with the second domain.
8. The system of claim 1 , wherein the one or more labels comprise one or more sublabels.
9. The system of claim 1 , wherein a subject matter of the first domain is different than a subject matter of the second domain.
10. A computerized method in which unstructured computer text is analyzed for classification and routing using artificial intelligence transfer learning, the method comprising: storing, in a computer data store, a first corpus of unstructured computer text associated with a first domain and a second corpus of unstructured computer text associated with a second domain; filtering, by a server computing device in communication with the computer data store, the first corpus of unstructured computer text and the second corpus of unstructured text using natural language processing; generating, by the server computing device, a first vocabulary data set based upon the filtered first corpus of unstructured computer text; generating, by the server computing device, a second vocabulary data set based upon the filtered second corpus of unstructured computer text; generating, by the server computing device, a one-hot vector for each word in the first vocabulary data set and the second vocabulary data set; training, using an artificial intelligence neural network executing on the server computing device, a word embedding model using the one-hot vectors of word pairs from the filtered first corpus of unstructured computer text and the filtered second corpus of unstructured computer text; creating, by the server computing device, a multidimensional vector for each word in the filtered first corpus of unstructured computer text and the filtered second corpus of unstructured computer text using the trained word embedding model; building, by the server computing device, a 3D tensor for the filtered second corpus of unstructured computer text using the trained word embedding model, the 3D tensor comprising a plurality of matrices, wherein each matrix corresponds to a sentence in the filtered second corpus of unstructured computer text and each matrix comprises a plurality of multidimensional vectors, each multidimensional vector corresponding to a word in the sentence; applying, by the server computing device, one or more labels to the plurality of matrices in the 3D tensor; training, using a recurrent artificial intelligence neural network executing on the server computing device, a long short-term memory model using the plurality of matrices in the 3D tensor and the corresponding labels; receiving, by the server computing device, a message comprising one or more sentences of unstructured computer text associated with the second domain from a first client computing device; filtering, by the server computing device, the unstructured computer text in the message using natural language processing; generating, by the server computing device, a matrix for each sentence in the unstructured computer text in the message by applying the trained word embedding model to each word in the sentence; identifies, by the server computing device, one or more labels, and a probability for each label, for each sentence in the unstructured computer text in the message by applying the trained long short-term memory model to the generated matrix for each sentence; assigning, by the server computing device, a label having a highest probability to each sentence in the unstructured computer text; and routing, by the server computing device, the message to a second client computing device based upon the assigned label.
11. The method of claim 10 , wherein filtering the first corpus of unstructured computer text and the second corpus of unstructured text using natural language processing comprises one or more of: (i) removing stopwords from the first corpus of unstructured computer text and the second corpus of unstructured text, (ii) lemmatizing the first corpus of unstructured computer text and the second corpus of unstructured text, or (iii) removing one or more symbols or digits from the first corpus of unstructured computer text and the second corpus of unstructured text.
12. The method of claim 10 , wherein the first corpus of unstructured computer text and the second corpus of unstructured computer text are input into the computer data store via a web page, input directly into the computer data store via a first computer file, input into the computer data store via a data feed, or any combination thereof.
13. The method of claim 10 , wherein generating a first vocabulary data set based upon the filtered first corpus of unstructured computer text comprises, for each word in the filtered first corpus of unstructured computer text: determining a number of times that the word appears in the filtered first corpus of unstructured computer text, and adding the word to a first vocabulary data set if the number of times that the word appears in the filtered first corpus of unstructured computer text is above a predetermined threshold.
14. The method of claim 10 , wherein generating a second vocabulary data set based upon the filtered second corpus of unstructured computer text comprises, for each word in the filtered second corpus of unstructured computer text: determining a number of times that the word appears in the filtered second corpus of unstructured computer text; and adding the word to a second vocabulary data set if the number of times that the word appears in the filtered second corpus of unstructured computer text is above a predetermined threshold.
15. The method of claim 10 , wherein the artificial intelligence neural network that trains the word embedding model comprises a shallow neural network having an input layer, a hidden layer, and an output layer.
16. The method of claim 10 , wherein the first corpus of unstructured computer text associated with the first domain is larger than the second corpus of unstructured computer text associated with the second domain.
17. The method of claim 10 , wherein the one or more labels comprise one or more sublabels.
18. The method of claim 10 , wherein a subject matter of the first domain is different than a subject matter of the second domain.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
May 31, 2018
June 9, 2020
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.